【作者】 王洪希；

【导师】 杨卫东；

【作者基本信息】 北京科技大学 ， 控制科学与工程， 2015， 博士

【摘要】 轧辊偏心是影响现代高精度板带轧制厚度质量的关键因素之一,要想进一步提高板带轧机的板形及厚度质量,必须对轧辊偏心扰动加以抑制及补偿控制。由于轧辊偏心信号是混杂在各种扰动和随机信号之中的复杂高频周期信号,轧辊偏心进行补偿控制的效果取决于如何从复杂的轧制力信号中准确提取出微弱的偏心信号,即轧辊偏心信号准确提取是偏心补偿控制的关键。为此,本课题针对现有各类轧辊偏心信号提取方法的局限性,提出采用现代空间谱估计中的子空间分解类算法来研究轧辊偏心信号提取问题,并对该类算法进行了融合与改进,通过理论分析、仿真及实验研究相结合来验证提取轧辊偏心信号的有效性和精确性,该类算法尤其在频率分辨率和抗噪声两个方面要比FFT法具有较好的优越性。论文工作的主要创新点和研究成果如下：1)针对FFT法在轧辊偏心提取中存在频率分辨率低且消噪效果不佳的局限性,本文提出改进型的噪声子空间MUSIC算法应用于轧辊偏心信号提取。从应用技术创新的层面,重点研究基于Root-MUSIC法和Prony法相融合的轧辊偏心信号提取新方法。仿真结果验证了方法结合的有效性,分辨率高、抗噪性强。2)为了减小计算量,提高算法的实时性,便于实际工程的应用。本文提出改进型的信号子空间ESPRIT算法应用于轧辊偏心信号提取。基于信号子空间的ESPRIT算法与基于噪声子空间的MUSIC法相比,不再考虑信号子空间与噪声子空间之间的关系,无需进行谱峰搜索,为此提高了算法的实用性。仿真分析出三种ESPRIT方法在不同阵元数和不同信噪比下的频率估计性能。3)针对轧辊偏心信号实际可能存在非平稳性质和测量噪声有色的性质,经典的二阶统计量子空间类算法存在有偏性和非一致性的问题,本文提出高阶累计量MUSIC法和Prony法相融合并应用于轧辊偏心信号提取。采用基于高阶累积量的MUSIC法对偏心信号进行空问分解达到降阶的效果,能够有效地抑制噪声,在信噪比低时仍具有高的频谱分辨率,能准确提取出偏心谐波的频率及谐波的个数。然后使用Prony方法进一步估计偏心信号的各次谐波幅值和相位,弥补了Prony法对噪声的敏感的弱点。仿真结果验证了该融合方法的有效性。以某厂热轧生产线为背景,采用现场轧制力数据,验证了基于高阶累计量MUSIC和Prony融合方法提取轧辊偏心信号具有很好的实际效果,实验结果表明该方法能准确地提取了相近频率成分及高次偏心谐波分量参数,且去噪效果明显,使重构偏心信号的精度很高,偏心补偿后的效果明显优于FFT法。更多还原

【Abstract】 Rolling eccentricity is part of the key factors affecting the thickness of strip quality in modern high-precision rolling process. In order to further enhance the strip mill quality of flatness and thickness, an effective compensation control and disturbances rejection in rolling eccentricity must be applied. Due to roll eccentricity signal is mixed in a complex high-frequency periodic signal, in which various disturbances surrounding. The control effect of rolling eccentricity compensation depends on how to accurately extract the weak eccentricity signal from the complex rolling force signal that the accurate signal extraction algorithm is a critical factor in eccentricity compensation control. To this point, a summarization of research on limitations of the existing extraction methods for rolling eccentricity signal is made. And it is proposed that apply subspace decomposition algorithm in the contemporary spatial spectrum estimation to solve the rolling eccentricity signal extraction problem is validity, and as well as related integration and improvement of this class of algorithm is given. Through a combination of theoretical analysis, simulation and experimental study to validate the efficiency and accuracy of this signal extraction algorithm, especially with this algorithm, the performances of the frequency resolution and noise suppression are much better than the FFT method. The main innovations in this research thesis are as follows:First,for the limitations of low frequency resolution and poor noise-canceling effect of the FFT method in rolling eccentricity extraction, it is proposed that the improved version of noise subspace MUSIC algorithm apply to the eccentricity signal extraction. From the aspect of technological innovation for application, focusing on a novel extraction algorithm based on the fusion of Prony and Root-MUSIC methods. Simulation results show the effectiveness of the new algorithm combining of high resolution, robust noise immunity.Second,in order to decrease the computational burden, to improve the online timeliness, and to ease of application in actual industrial tasks. This is such an original idea that an improved subspace ESPRIT algorithm is applied to roll eccentricity signal extraction. The subspace ESPRIT algorithm has more usability compared to the MUSIC method, for it no longer examines the relationship between the signal subspace and noise subspace, no spectral peak searching. Simulation for analysing the frequency estimation performance of three different ESPRIT methods is given under different number of array elements and different SNR.Third,to the characteristics of rolling eccentricity signal that the measurement is non-stationary and colored noise is actually existed, and the nature of classical algorithm of second-order statistics in quantum space has the problems of bias and non-uniformity. So the integrating method with both higher-order cumulant MUSIC and Prony is used to eccentricity signal extraction. Based on this method, spatial decomposition of eccentricity achieves reduced-order effect. This method still has a high spectral resolution and can accurately extract the frequencies of the eccentric harmonics and the number of the harmonics. Then Prony method is used to estimate the amplitudes and phases of the harmonics, which remedies the weaknesses of sensitive to noise and the demand for the order of the signal. The simulation shows that combining the two ways is effective.Taking a hot-rolled production line in a factory as a background, the proposed method is verified by using the rolling force data, and by the MUSIC algorithm based on HOC and Prony method to extract rolling eccentricity signal. The experimental results demonstrate that the method can accurately extract parameters of similar frequency and high-order harmonics. And the proposed reconstruction eccentric model has higher precision than the FFT method. The eccentricity compensation effect is obvious.更多还原